Well to put things into context Unreal 4 consists of feature sets based on how things use to work that also combines feature sets that is in place to contend with what could be done via research and discovery as to the need for much more complex movement systems.
What seems to be the case is the need for the straight ahead bottom up design pathway where the mind map says it needs to work today instead of building up a frame work which is the need between code and content of just doing enough to make it work versus the required R&D which is usually considered a time consuming process as to proof of concept and time spent with real time development usually = $$$$$
That being said state machines are very easy to understand as to animation migration as to a solution that has been proven to work and every game since the beginning of game development history has used some kind of state machine that requires a nested argument built into the pathway based on the need to change a animation state within a parallel animation migration pathway.
In most cases state machines are still used today as they are off the shelf solutions that works well in combination with the movement component as to the need for ready made replication requirements.
The back end is the tendency for a state machine to become very sloppy that adds input latency from the input device giving it that kind of sluggish feeling as the design scales up as to the need of state changes.
The direction with animation needs in Unreal 4 though seems to be more towards data driven top down migration pathways that excludes the need for argument based state machines starting with the locomotion state and layering animation state changes with out the use of a state machine.
To avoid the 10 page design documentation I noticed like many that over time state machines became more and more difficult to understand to even be able to get out of fixed state and over time my conclusion is the only practical reason to included a state machine is when you don’t want the player to leave the state as to cause and effect.
So to answer the question with a question.
Do you even need to use a state machine at all?
In my case the answer is no but requires a development pathway that makes assumptions as to what might be needed down the road instead of what works today in exchange for top down design ideology in an effort to build the framework which is what the state machine is as an off the shelf ready to use solution.
With that in mind watch Epic’s twitch casts of the animation tech used in Paragon and the ah ha moment is that the animation migration is 100% data driven and the question is where is the data coming from?